Initial Deployment of Distributed Java Programs in Clusters of JVMs Through Extremal Optimization Approach

The paper presents an algorithm for initial Java program deployment based on the extremal optimization approach. a Java program is represented as a DAG in which tasks are representing methods of Java objects in distributed programs which meet the requirements of the ProActive Java execution framework. The extremal optimization is used for DAG scheduling in a system of Java Virtual Machines communicating using a RMI mechanism implemented in a cluster of multicore processors. Extremal optimization is designed based on two local fitness functions in which reduction of delays of task execution after reception of data and task execution time unbalance in processors are used as heuristics for improvements of extremal optimization solutions.